Overview:
Minutes is an open-source conversation memory tool designed to capture the human side of interactions—decisions, intent, and context—and make it queryable by AI agents. It runs entirely locally and writes every meeting as plain Markdown files to a ~/meetings/ directory. The tool is built for developers, AI agents (like Claude Code, Codex, OpenCode, and Gemini CLI), and users who want a durable, file-based repository of their conversations. It supports local audio capture, transcription, voice memo processing, and a searchable relationship graph that tracks people, commitments, and topics across all meetings.
Core Features:
Conversation memory for agents: Exposes meeting artifacts as plain Markdown files that AI tools like Claude Code and Gemini CLI can read directly without an SDK or API key.
Local audio capture & transcription: Records meetings and voice memos using your machine's microphone, with local speaker diarization and transcription via whisper.cpp. Audio never leaves your machine unless you opt into cloud summarization.
Search & relationship intelligence: Tracks people, commitments, and topics across all meetings via a SQLite index. Supports queries like "what did I promise Sarah?".
Multi-surface access: Provides a desktop app (Tauri), CLI, MCP server, and OpenCode project integrations for recording, searching, and managing meeting memory.
Automation & knowledge base: Emits scheduled summaries, context bundles, and structured knowledge files (person profiles, decisions) that accumulate over time, supporting Wiki, PARA, and Obsidian output formats.
Use Cases:
Developers who want their AI coding agents to recall past decisions, commitments, and context from meetings without relying on cloud services.
Users needing a searchable, locally-stored archive of conversations from various sources (Zoom, Google Meet, phone voice memos).
Teams using AI coding agents (like Claude Code or Codex) that require a durable, file-based memory layer for project context and relationship tracking.
Self-hosters looking for a completely local alternative to cloud meeting note-takers that writes data to plain files, ensuring future accessibility via basic tools like
grep.
Why It Matters:
Minutes differentiates itself by writing every conversation to open, plain Markdown files rather than a proprietary database or cloud service. This approach means your conversation archive remains directly readable by any AI or standard tool (like grep) without vendor dependency, SDKs, or API keys. The file-based architecture is designed as "conversation infrastructure for agents," making it extensible via MCP, CLI, and a TypeScript SDK. The default configuration is local-only, with speaker diarization and transcription processed on your own hardware.
